Chern-Weil theory gives characteristic classes (e.g. Chern class, Euler class, Pontryagin) of a vector bundle in terms of polynomials in the curvature form of an arbitrary connection. There seems to be no hope in getting Stiefel-Whitney classes from this method since Chern-Weil gives cohomology classes with real coefficients while Stiefel-Whitney classes have $\mathbb Z/2$ coefficients. Further, since any vector bundle over a curve has vanishing curvature, classes obtained by Chern-Weil can't distinguish, for example, the Mobius bundle from the trivial bundle over the circle (while Stiefel-Whitney classes do).

Nonetheless, I am wondering if there is a more general or abstract framework that allows one to define the Stiefel-Whitney classes in the spirit of Chern-Weil. For example, maybe this is done through a more abstract definition of a connection/curvature.

  • $\begingroup$ Might be worth looking nLabwards, e.g., ncatlab.org/nlab/show/… or ncatlab.org/nlab/show/… or math.ntnu.no/~stacey/Mathforge/nForum/… . $\endgroup$ Jan 18, 2011 at 17:06
  • $\begingroup$ I don't have an answer to this question, but I have a guess. In general, if you want to use De Rham theory to recover integral invariants of a manifold, a smart thing to do is look at distributional differential forms (I first learned of the literature on this from MO, but I can't remember where). So I would start by defining a distributional connection on a vector bundle $E$ (rather, on its frame bundle) to be a distributional $E$-valued 1-form and going from there. $\endgroup$ Jan 18, 2011 at 18:52
  • 2
    $\begingroup$ You might want to take a look at "Stiefel-Whitney currents" by Harvey and Zweck [MR1731064 (2001c:53104)]. $\endgroup$
    – 680
    Jan 19, 2011 at 12:07
  • 1
    $\begingroup$ @David and 680: thanks for the links! $\endgroup$ Jan 20, 2011 at 23:37
  • 2
    $\begingroup$ @diverietti: I think the confusion is about real curves as opposed to complex curves ;-) $\endgroup$ Apr 11, 2011 at 14:41

2 Answers 2


I want to say the short answer is no.

But in certain contexts, we can get things that are analogous. For example if you take a principal $B$-bundle $Q$ over $M$ and then suppose you can have a "nice" :) central extension of your lie group $B$ by $$1 \to \underline{\mathbb{C}^*}_M \to \tilde{B} \to B \to 1,$$where $\underline{\mathbb{C}^*}_M$ is the sheaf of smooth functions into $\mathbb{C}^*$, then you can define a cohomology class in $H^1(M, \underline{B})$ by seeing how well you can lift your bundle to a $\tilde{B}$-bundle. Now by the central extension, we would have $$H^1(M, \underline{B}) \overset{\sim}{=} H^2(M , \underline{\mathbb{C}^*}_M)$$ and then by the exponential sequence you would have $$H^1(M, \underline{B}) \overset{\sim}{=} H^2(M , \underline{\mathbb{C}^*}_M) \overset{\sim}{=} H^3(M, \mathbb{Z}).$$

And so what I am saying (since principal bundles have associated vector bundles) that your vector bundle in the right conditions could give you a degree three cohomology class in the integers. I'm sure you could pluck your Z/2 coefficients out of this (I don't know why you would want to be so rigid, nor am I claiming you are asking for that). Then there is an actual geometric interpretation of this integer related to a certain curvature form in this construction that I am not quite ready to add to this answer yet :)

  • $\begingroup$ By the way, I'm fairly certain that Weil developed some of these methods. $\endgroup$
    – cheyne
    Feb 26, 2013 at 15:42

There exist flat manifolds (i.e., closed Riemannian manifolds with vanishing sectional curvature) which do not admit any spin or spin${}^c$-structure. But since the existence of spin, resp. of spin${}^c$-structures is detected by the second Stiefel-Whitney class, I am strongly led to believe that there is no way of defining these à la Chern-Weil. The reason is that the phrase 'à la Chern-Weil' entails for me that you use in an essential way the curvature tensor, but in these examples it vanishes.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.